Generating adversarial inputs for a graph neural network model of AC power flow
Research has been conducted to formulate and solve optimization problems aimed at generating adversarial inputs for a graph neural network model of AC power flow. This study specifically demonstrates the capability on the CANOS-PF model using the PFΔ benchmark library, revealing significant errors in predicted AC power flow solutions on a 14-bus test grid.
WPN Brief
- What Happened
Research has been conducted to formulate and solve optimization problems aimed at generating adversarial inputs for a graph neural network model of AC power flow. This study specifically demonstrates the capability on the CANOS-PF model using the PFΔ benchmark library, revealing significant errors in predicted AC power flow solutions on a 14-bus test grid.
- Why It Matters
The findings highlight the importance of developing rigorous verification and robust training methods for neural network surrogate models, which could enhance the reliability and accuracy of AC power flow predictions in practical applications.